
First attempt at making a lora through Civitai and publishing it. Forgive the scuff, both in this post and also in the lora. One thing is a lot of the training data had some simple backgrounds and text. I used 80 images to make it, 8 epochs seemed to be about right. Clip skip 1.
Suggested Positives: horrhist, round pupils, small pupils, black eyes, 2d, flat color
More Optional Positives IMO: upper body, solo, 1boy/1girl, male/female focus, male/female, smug/angry/light smile, medium breasts, indoors/outdoors, detailed background
Suggested Negatives: no pupils, large pupils, text, 3d, simple background, white background
Does good with upper body images, eyes can be difficult to get right especially with full body.
Would love to see if y'all can make some interesting art with this. I loved Horrible Histories growing up.
WORKS SURPRISINGLY WELL!!! I'd HIGHLY suggest using the tag "horrhist" (basically the trigger word) and then lora weight = 1.5 (lower weights like 1.2 can still give a nice and different result, but from my tests 1.5 is the most true to Horrible Histories.)
描述:
训练词语: horrhist
名称: 1090418_training_data.zip
大小 (KB): 16293
类型: Training Data
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success
名称: Horrible_Histories_Artstyle-000008.safetensors
大小 (KB): 223109
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success